Improvement of HVSR technique by cluster analysis

The Horizontal to Vertical Spectral Ratio (HVSR) technique, applied to ambient noise, is widely used to quickly estimate the fundamental frequency of a site. The HVSR technique is based on numerous assumptions about the propagation medium and the nature of the seismic noise, often hard to verified. In addition, in order to obtain reliable results, several acquisition and processing criteria must be respected. One of the most controversial aspects in the technique implementation are the reliable criteria for the identification in the microtremors signals of time windows appropriate for the calculation of the HVSR curves. Several authors suggest to remove spikes and transients because they bring information highly dependent from the source and cannot be used to estimate the resonance frequency of the site. Other authors instead, suggest that transients in seismic noise carry essential information and therefore should not be removed. The time windows suitable for the HVSR techniques are generally arbitrarily identified by the operator by a simple visual inspection of the signals in time or spectral domain. This can lead to an incorrect determination and interpretation of the HVSR curves. This technique is often used in the first level seismic microzonation. An incorrect processing of the seismic noise may provide unreliable results and HVSR curves difficult to interpret, an so making hard the identification of areas with similar seismic response. In this work we used the cluster analysis to analyze the HVSR curves as function of time and space. Cluster analysis is generally used to group objects characterized by a high level of similarity. We test different clustering algorithms to seek inside the seismic noise, time windows characterized by very similar HVSR curves. The implemented algorithms were applied to about 500 records of seismic noise each 46 minutes long. This analysis allowed us to identify inside each record, groups of very similar HVSR curves, allowing us to easily split those linked to the site effects from those related to the source effects. The separation of the two effects allowed us to determine robust HVSR curves better related to the local site effects. The clustering techniques were used also to group HVSR determined for different sites and identify areas with very similar behaviour in seismic perspective.